[PATCH v2] winegstreamer: Ensure that the source reports both PCM and floating point audio formats.

[Zebediah Figura (she/her)]

On 5/13/21 2:53 AM, Nikolay Sivov wrote:
> 
> 
> On 5/13/21 10:47 AM, Giovanni Mascellani wrote:
>> Hi,
>>
>> Il 13/05/21 09:18, Nikolay Sivov ha scritto:
>>> On 5/12/21 6:37 PM, Giovanni Mascellani wrote:
>>>> diff --git a/dlls/winegstreamer/media_source.c
>>>> b/dlls/winegstreamer/media_source.c
>>>> index eb5b9e366ec..5249bc9c41d 100644
>>>> --- a/dlls/winegstreamer/media_source.c
>>>> +++ b/dlls/winegstreamer/media_source.c
>>>> @@ -806,6 +806,31 @@ static HRESULT media_stream_init_desc(struct
>>>> media_stream *stream)
>>>>                    goto done;
>>>>            }
>>>>        }
>>>> +    else if (format.major_type == WG_MAJOR_TYPE_AUDIO)
>>>> +    {
>>>> +        /* Expose at least one PCM and one floating point type for the
>>>> +           consumer to pick from. */
>>>> +        static const enum wg_audio_format audio_types[] =
>>>> +        {
>>>> +            WG_AUDIO_FORMAT_S16LE,
>>>> +            WG_AUDIO_FORMAT_F64LE,
>>>> +        };
>>> Is F64LE correct? I thought before you used an equivalent of F32LE, if
>>> that's what (Float, MF_MT_AUDIO_BITS_PER_SAMPLE=32) means.
>>
>> Ugh, I meant F32LE. My mistake.
>>
>>>> +
>>>> +        stream_types = malloc( sizeof(IMFMediaType *) *
>>>> (ARRAY_SIZE(audio_types) + 1) );
>>>> +
>>>> +        stream_types[0] = mf_media_type_from_wg_format(&format);
>>>> +        type_count = 1;
>>>> +
>>>> +        for (i = 0; i < ARRAY_SIZE(audio_types); i++)
>>>> +        {
>>>> +            struct wg_format new_format;
>>>> +            if (format.u.audio.format == audio_types[i])
>>>> +                continue;
>>>> +            new_format = format;
>>>> +            new_format.u.audio.format = audio_types[i];
>>>> +            stream_types[type_count++] =
>>>> mf_media_type_from_wg_format(&new_format);
>>>> +        }
>>>> +    }
>>>>        else
>>>>        {
>>>>            stream_type = mf_media_type_from_wg_format(&format);
>>> Thanks, that looks fine to me. It could be reduced a bit more still, by
>>> removing 'new_format' for example, but that's minor.
>>
>> That would be wrong, wouldn't it? If I change format.u.audio.format,
>> then the "if () continue" at the beginning of the loop doesn't make
>> sense any more. Unless I remember the original format in another
>> variable, which would then have comparable complexity.
>>
>>>    Now we need someone
>>> who knows gstreamer to comment, specifically if resampling PCM ->
>>> PCM/float -> float would work. Or maybe you can test that yourself with
>>> some sin wave data with rate not matching output device rate.
>>
>> I can't see how this is related to my patch: I don't think Gstreamer
>> does any rate change at this level. It just advertises the native rate
>> the audio stream supports. Whatever rate matching is required, it will
>> be done downstream from the media source (both before and after my
>> patch, which is only concerned with the sample format, not its rate).
> Original patch added PCM->F32LE and float->S16LE conversions, this one
> also adds conversions between different PCM and different float formats.
> So the question is does it work? It was wrong to call it rate, I meant
> bps of course, or format as whole.

audioconvert should be perfectly capable of this, yes.